Title: A pipeline for Mendelian randomization studies using large metabolomics data as intermediates.

Start date: 01/10/2017

Expected submission date: 01/04/2021

Aim

Identify metabolites that sit on the causal pathway from increased adiposity to disease

Objectives

1. Identify diseases causally associated with increased adiposity
2. Identify and describe appropriate instrumentation of increased adiposity
3. Identify metabolites causally associated with increased adiposity
4. Compare and implement methods and rules to cluster metabolites
5. Identify diseases causally associated with metabolites

Can we build a flowchart of decisions to achieve this?

Chapter 1: overview

Increased adiposity and associtions with disease

Chapter (50%): writing (50%), analysis (100%)

Chapter gives backgraound on problem and context of thesis

• What is the problem
  • overweight = ~40% (EU = ~60%)
  • obese = ~13% (EU = ~23%)
• What is adiposity
  • adipose cells - idea of genetics and metabolites introduced
  • different measures
• Broad overview of disease
  • what diseases are asscoiated with increased adiposity
  • semi-systematic way of doing this without performing systematic review(?)
• Application of metabolites
  • adipose tissue is a signalling organ
• Introduce MR

Chapter 1: identify diseases associated with increased adiposity

MELODI analysis

• 975,402 articles
• 10,828 enriched terms
• Filtering left 77 terms
• 8 categories plus ‘Other’

Chapter 1: MELODI output

Chapter 2: overview

Systematic review

Chapter (40%): writing (40%), analysis (40%)

Paper (20%): writing (20%), analysis (40%)

What has MR told us about the causal relevance of increased adiposity?

Inclusion: All MR studies using a measure of increased adiposity as the exposure

• Literature search (100%)
• Screening (100%)
• Data extraction (30%)
  • 60 papers
  • ~800 exposure-outcome analyses
• Writing (40%)

• Meta-analysis - may not be possible
  • Lots of differences between studies (e.g. different ancestries, different methods, different SNPs)
• Reporting is poor
  • GWAS reporting is a problem - e.g. multiple GWASs in one paper but only one discussed

Chapter 3: overview

Instrumenting measures of increased adiposity for MR analyses

Chapter (5%): writing (5%), analysis (10%)

Paper: form part of systematic review paper

Chapter presents current practices for instrumenting BMI, WHR and BF% in MR analyses and looks at associations of instruments with potential confounders

• Current instrumentation practices (will get from SR)
  • SNP lists most common etc.
• GWAS descriptions
  • what GWASs are used - e.g. does everyone use Locke 2015 for BMI
• Instruments
  • analysis
    • associations with exposures (e.g. BMI SNPs with measured WHR)
    • instrument strength
    • instruments and potential confounders
      • metabolites: age, sex, smoking status
      • diseases: age, sex, smoking status

F statistics for 4 meausres of adiposity (BF%, BMI, WHR, WHRadjBMI) across multiple SNP sets. Mean given as black diamond and blue line indicating an F statistic of 10

Chapter 4: overview

Observational analysis: increased adiposity and metabolites

Chapter (10%): writing (5%), analysis (20%)

Paper: combined with chapter 5? / favourable adiposity paper?

Chapter explores observational associations between measures of increased adiposity and metabolites

• Exposures
  • BMI / BMI GRS
  • BF% / favourable adiposity GRS
• 230 metabolites
• ALSPAC mothers
  • 1 time point / 2 clinics
• ALSPAC kids
  • 3 time points (7, 12, 18)
  • sex combined
  • male / female
• Expand to WHR / WHRadjBMI
• Replication - INTERVAL / FGFP

Chapter 5: overview

MR analysis: increased adiposity and metabolites

Chapter (40%): writing (50%), analysis (90%)

Paper (40%): writing (50%), analysis (90%)

• Exposures
  • BMI: Yengo 2018, 941 SNPs (conditional on joint analysis)
  • WHR: Pulit 2018, 316 SNPs
  • BF%: Lu 2016, 7 SNPs
• 3 exposures, 123 metabolites
  • 369 tests
  • sensitivity analysis = 1107 tests
• Additional sensitivity analyses = 7380
  • BMI: Yengo not-COJO 656 SNPs; BMI not biobank 77 SNPs
  • WHR: Shungin 28 SNPs
  • BF%: Lu 5 SNPs, Hubel (biobank) 76 SNPs
  • WHRadjBMI Pulit 346 SNPs
  • Clumped versions of all (exc. Lu)
• Total tests = 8856

• Additional metabolite data
  • Shin metabolites = 452
    • Total tests = 2.4408^{4}
  • Unlikely to include in thesis

Chapter 5: main figure

Chapter 5: main figure as forest plot

Chapter 5: “significant” associations forest plot

Chapter 5: consistent directions of effect

  exposure1 exposure2       direction
1         1         1 Positive effect
2        -1        -1 Negative effect
3         1        -1 Opposite effect
4        -1         1 Opposite effect

Chapter 5: consistent “significant” associations

• Phenylalanine
  • Phenylalanine is an essential amino acid
  • If BMI etc. is higher does this correspond to an increase in consumption of phenylalanine food sources?
• Tyrosine
  • Phenylalanine is converted into tyrosine in the body
  • If increased consumption of phenylalnine (as above) would this lead to a corresponding increase in tyrosine?
• Apolopoprotein A1 (ApoA1)
  • ApoA1 is a major component of HDL particles in plasma
  • Does an increase in BMI etc. and subsequent increase in LDL particles lead to a reduction in HDL particles and thus a reduction in Apo A1?
  • OR, is the reduction in Apo A1 the cause of the increase in LDL particles?

Chapter 6: overview

MR Viz / EpiViz / EpiCirocs / Mulit Variable Viz / MAVIS: Multi vAriable VisualISation….

Chapter (50%): writing (50%), analysis (90%)

Paper (50%): writing (50%), analysis (100%)

R pakcage (90%)

Web app (90%)

Chapter describes the problem with analyses such as that in chapter 5 and how global overview is required and presents a tool for this

• Web app (90%)
  • Streamlines process of making Circos plots
  • No coding skills required - R code is extensive - remvoes barrier
  • Limited customisation
• R package (90%)
  • Streamlines process while allowing advanced features
  • Basic coding skills required
  • Extensive customisation

Chapter 6: R code

## function (track_number, track1_data, track2_data, track3_data, 
##     track1_type, track2_type, track3_type, label_column, section_column, 
##     order = TRUE, order_column, estimate_column, pvalue_column, 
##     pvalue_adjustment, lower_ci, upper_ci, lines_column, lines_type = "o", 
##     bar_column, histogram_column, histogram_binsize = 0.01, histogram_densityplot = FALSE, 
##     legend = FALSE, track1_label = NA, track2_label = NA, track3_label = NA, 
##     pvalue_label = NA, circle_size = 25, track1_height = 0.2, 
##     track2_height = 0.2, track3_height = 0.2) 
## {
##     track1 <- 1
##     track2 <- 2
##     track3 <- 3
##     track4 <- 4
##     x_axis_index <- 1
##     track_axis_reference <- 0
##     margins <- c(0.5, 0.5, 0.5, 0.5) * 25
##     start_gap <- 17
##     start_degree <- 90
##     section_track_height <- 0.1
##     discrete_palette <- c("#00378f", "#ffc067", "#894300")
##     section_fill_colour <- "snow2"
##     section_text_colour <- "black"
##     section_line_colour <- "grey"
##     section_line_thickness <- 1.5
##     section_line_type <- 1
##     reference_line_colour <- "deeppink"
##     reference_line_thickness <- 1.5
##     reference_line_type <- 1
##     point_pch <- 21
##     point_cex <- 1.5
##     point_col1 <- discrete_palette[1]
##     point_bg1 <- "white"
##     point_col1_sig <- "white"
##     point_bg1_sig <- discrete_palette[1]
##     point_col2 <- discrete_palette[2]
##     point_bg2 <- "white"
##     point_col2_sig <- "white"
##     point_bg2_sig <- discrete_palette[2]
##     point_col3 <- discrete_palette[3]
##     point_bg3 <- "white"
##     point_col3_sig <- "white"
##     point_bg3_sig <- discrete_palette[3]
##     ci_lwd <- 5
##     ci_lty <- 1
##     ci_col1 <- discrete_palette[1]
##     ci_col2 <- discrete_palette[2]
##     ci_col3 <- discrete_palette[3]
##     lines_col1 <- discrete_palette[1]
##     lines_col2 <- discrete_palette[2]
##     lines_col3 <- discrete_palette[3]
##     lines_lwd <- 3
##     lines_lty <- 1
##     y_axis_location <- "left"
##     y_axis_tick <- FALSE
##     y_axis_tick_length <- 0
##     y_axis_label_cex <- 0.75
##     label_distance <- 1.5
##     label_col <- "black"
##     label_cex <- 0.6
##     data <- track1_data
##     if (order == TRUE) {
##         data <- data[order(data[[section_column]], data[[label_column]]), 
##             ]
##     }
##     else if (order == FALSE) {
##         data[[section_column]] <- stats::reorder(data[[section_column]], 
##             data[[order_column]])
##         data <- data[order(data[[section_column]], data[[label_column]]), 
##             ]
##     }
##     data$x <- with(data, ave(seq_along(data[[section_column]]), 
##         data[[section_column]], FUN = seq_along))
##     npercat <- as.vector(table(data[[section_column]]))
##     getaxis <- function(data) {
##         for (i in 1:nrow(data)) {
##             data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                 data[[section_column]][i])))
##             data$ncat[i] <- data$x[i]/data$n[i]
##         }
##         return(data)
##     }
##     data <- getaxis(data)
##     data$section_numbers = factor(data[[section_column]], labels = 1:nlevels(data[[section_column]]))
##     gap = c(rep(1, nlevels(data[[section_column]]) - 1), start_gap)
##     circlize::circos.clear()
##     graphics::par(mar = c(0.6, 0.5, 0.5, 0.5) * circle_size, 
##         cex = 0.8, xpd = NA)
##     circlize::circos.par(cell.padding = c(0, 0.5, 0, 0.5), start.degree = start_degree, 
##         gap.degree = gap, track.margin = c(0.012, 0.012), points.overflow.warning = FALSE, 
##         track.height = section_track_height, clock.wise = TRUE)
##     circlize::circos.initialize(factors = data$section_numbers, 
##         xlim = c(0, 1), sector.width = npercat)
##     circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##         track.index = track1, x = data$ncat, ylim = c(0, 1), 
##         track.height = 0.05, panel.fun = function(x, y) {
##             chr = circlize::get.cell.meta.data("sector.index")
##             xlim = circlize::get.cell.meta.data("xlim")
##             ylim = circlize::get.cell.meta.data("ylim")
##             circlize::circos.rect(xlim[1], 0, xlim[2], 1, border = NA, 
##                 col = section_fill_colour)
##             circlize::circos.text(mean(xlim), mean(ylim), chr, 
##                 cex = 1, facing = "outside", niceFacing = TRUE, 
##                 col = section_text_colour)
##         }, bg.border = NA)
##     circlize::circos.trackText(factors = data$section_numbers, 
##         track.index = track1, x = data$ncat, y = data$b * 0 + 
##             label_distance, labels = data[[label_column]], facing = "reverse.clockwise", 
##         niceFacing = TRUE, adj = c(1, 1), col = label_col, cex = label_cex)
##     if (track_number >= 1 && track1_type == "points") {
##         data <- track1_data
##         track.index <- 2
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         a <- min(data[[lower_ci]])
##         b <- min(data[[upper_ci]])
##         axis_min <- min(a, b)
##         axis_min <- round(axis_min, 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         a <- max(data[[lower_ci]])
##         b <- max(data[[upper_ci]])
##         axis_max <- max(a, b)
##         axis_max <- round(axis_max, 3)
##         axis_max <- round(axis_max + (axis_max * 0.01), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         for (i in 1:nlevels(data$section_numbers)) {
##             data1 = subset(data, section_numbers == i)
##             circlize::circos.trackPlotRegion(factors = data1$section_numbers, 
##                 track.index = track.index, x = data1$ncat, y = data1[[estimate_column]], 
##                 ylim = c(axis_min, axis_max), track.height = track1_height, 
##                 bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                   y) {
##                   circlize::circos.lines(x = x, y = y * 0 + track_axis_reference, 
##                     col = reference_line_colour, lwd = reference_line_thickness, 
##                     lty = reference_line_type)
##                   circlize::circos.segments(x0 = data1$ncat, 
##                     x1 = data1$ncat, y0 = data1[[estimate_column]] * 
##                       0 - -(data1[[lower_ci]]), y1 = data1[[estimate_column]] * 
##                       0 + data1[[upper_ci]], col = ci_col1, lwd = ci_lwd, 
##                     lty = ci_lty, sector.index = i)
##                 })
##         }
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] > 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] > pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] > pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col1, 
##             bg = point_bg1)
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] < 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] < pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] < pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col1_sig, 
##             bg = point_bg1_sig)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 1 && track1_type == "lines") {
##         data <- track1_data
##         track.index <- 2
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[lines_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[lines_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[lines_column]], 
##             ylim = c(axis_min, axis_max), track.height = track1_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col1, 
##                   lwd = lines_lwd, lty = lines_lty, type = lines_type)
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 1 && track1_type == "bar") {
##         data <- track1_data
##         track.index <- 2
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[bar_column]]), 3)
##         axis_max <- round(max(data[[bar_column]]), 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[bar_column]], 
##             ylim = c(axis_min, axis_max), track.height = track1_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col1, 
##                   lwd = 1, lty = lines_lty, type = "s", area = T, 
##                   border = "White", baseline = "bottom")
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 1 && track1_type == "histogram") {
##         data <- track1_data
##         track.index <- 2
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[histogram_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[histogram_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackHist(factors = data$section_numbers, 
##             x = data[[histogram_column]], track.height = track1_height, 
##             track.index = NULL, col = lines_col1, border = lines_col1, 
##             bg.border = NA, draw.density = histogram_densityplot, 
##             bin.size = histogram_binsize)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 2 && track2_type == "points") {
##         data <- track2_data
##         track.index <- 3
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         a <- min(data[[lower_ci]])
##         b <- min(data[[upper_ci]])
##         axis_min <- min(a, b)
##         axis_min <- round(axis_min, 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         a <- max(data[[lower_ci]])
##         b <- max(data[[upper_ci]])
##         axis_max <- max(a, b)
##         axis_max <- round(axis_max, 3)
##         axis_max <- round(axis_max + (axis_max * 0.01), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         for (i in 1:nlevels(data$section_numbers)) {
##             data1 = subset(data, section_numbers == i)
##             circlize::circos.trackPlotRegion(factors = data1$section_numbers, 
##                 track.index = track.index, x = data1$ncat, y = data1[[estimate_column]], 
##                 ylim = c(axis_min, axis_max), track.height = track2_height, 
##                 bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                   y) {
##                   circlize::circos.lines(x = x, y = y * 0 + track_axis_reference, 
##                     col = reference_line_colour, lwd = reference_line_thickness, 
##                     lty = reference_line_type)
##                   circlize::circos.segments(x0 = data1$ncat, 
##                     x1 = data1$ncat, y0 = data1[[estimate_column]] * 
##                       0 - -(data1[[lower_ci]]), y1 = data1[[estimate_column]] * 
##                       0 + data1[[upper_ci]], col = ci_col2, lwd = ci_lwd, 
##                     lty = ci_lty, sector.index = i)
##                 })
##         }
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] > 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] > pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] > pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col2, 
##             bg = point_bg2)
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] < 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] < pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] < pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col2_sig, 
##             bg = point_bg2_sig)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 2 && track2_type == "lines") {
##         data <- track2_data
##         track.index <- 3
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[lines_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[lines_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[lines_column]], 
##             ylim = c(axis_min, axis_max), track.height = track2_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col2, 
##                   lwd = lines_lwd, lty = lines_lty, type = lines_type)
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 2 && track2_type == "bar") {
##         data <- track1_data
##         track.index <- 3
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[bar_column]]), 3)
##         axis_max <- round(max(data[[bar_column]]), 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[bar_column]], 
##             ylim = c(axis_min, axis_max), track.height = track2_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col2, 
##                   lwd = 1, lty = lines_lty, type = "s", area = T, 
##                   border = "White", baseline = "bottom")
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 2 && track2_type == "histogram") {
##         data <- track2_data
##         track.index <- 3
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[histogram_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[histogram_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackHist(factors = data$section_numbers, 
##             x = data[[histogram_column]], track.height = track2_height, 
##             track.index = NULL, col = lines_col2, border = lines_col2, 
##             bg.border = NA, draw.density = histogram_densityplot, 
##             bin.size = histogram_binsize)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 3 && track3_type == "points") {
##         data <- track3_data
##         track.index <- 4
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         a <- min(data[[lower_ci]])
##         b <- min(data[[upper_ci]])
##         axis_min <- min(a, b)
##         axis_min <- round(axis_min, 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         a <- max(data[[lower_ci]])
##         b <- max(data[[upper_ci]])
##         axis_max <- max(a, b)
##         axis_max <- round(axis_max, 3)
##         axis_max <- round(axis_max + (axis_max * 0.01), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         for (i in 1:nlevels(data$section_numbers)) {
##             data1 = subset(data, section_numbers == i)
##             circlize::circos.trackPlotRegion(factors = data1$section_numbers, 
##                 track.index = track.index, x = data1$ncat, y = data1[[estimate_column]], 
##                 ylim = c(axis_min, axis_max), track.height = track3_height, 
##                 bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                   y) {
##                   circlize::circos.lines(x = x, y = y * 0 + track_axis_reference, 
##                     col = reference_line_colour, lwd = reference_line_thickness, 
##                     lty = reference_line_type)
##                   circlize::circos.segments(x0 = data1$ncat, 
##                     x1 = data1$ncat, y0 = data1[[estimate_column]] * 
##                       0 - -(data1[[lower_ci]]), y1 = data1[[estimate_column]] * 
##                       0 + data1[[upper_ci]], col = ci_col3, lwd = ci_lwd, 
##                     lty = ci_lty, sector.index = i)
##                 })
##         }
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] > 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] > pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] > pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col3, 
##             bg = point_bg3)
##         circlize::circos.trackPoints(factors = subset(data, data[[pvalue_column]] < 
##             pvalue_adjustment)$section_numbers, track.index = track.index, 
##             x = subset(data, data[[pvalue_column]] < pvalue_adjustment)$ncat, 
##             y = subset(data, data[[pvalue_column]] < pvalue_adjustment)[[estimate_column]], 
##             cex = point_cex, pch = point_pch, col = point_col3_sig, 
##             bg = point_bg3_sig)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 3 && track3_type == "lines") {
##         data <- track3_data
##         track.index <- 4
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[lines_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[lines_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[lines_column]], 
##             ylim = c(axis_min, axis_max), track.height = track3_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col3, 
##                   lwd = lines_lwd, lty = lines_lty, type = lines_type)
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 3 && track3_type == "bar") {
##         data <- track1_data
##         track.index <- 4
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[bar_column]]), 3)
##         axis_max <- round(max(data[[bar_column]]), 3)
##         circlize::circos.trackPlotRegion(factors = data$section_numbers, 
##             track.index = track.index, x = data$ncat, y = data[[bar_column]], 
##             ylim = c(axis_min, axis_max), track.height = track3_height, 
##             bg.border = NA, bg.col = NA, panel.fun = function(x, 
##                 y) {
##                 circlize::circos.lines(x = x, y = y, col = lines_col3, 
##                   lwd = 1, lty = lines_lty, type = "s", area = T, 
##                   border = "White", baseline = "bottom")
##             })
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (track_number >= 3 && track3_type == "histogram") {
##         data <- track3_data
##         track.index <- 4
##         if (order == TRUE) {
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         else if (order == FALSE) {
##             data[[section_column]] <- stats::reorder(data[[section_column]], 
##                 data[[order_column]])
##             data <- data[order(data[[section_column]], data[[label_column]]), 
##                 ]
##         }
##         data$x <- with(data, ave(seq_along(data[[section_column]]), 
##             data[[section_column]], FUN = seq_along))
##         npercat <- as.vector(table(data[[section_column]]))
##         getaxis <- function(data) {
##             for (i in 1:nrow(data)) {
##                 data$n[i] <- as.numeric(nrow(subset(data, data[[section_column]] == 
##                   data[[section_column]][i])))
##                 data$ncat[i] <- data$x[i]/data$n[i]
##             }
##             return(data)
##         }
##         data <- getaxis(data)
##         data$section_numbers = factor(data[[section_column]], 
##             labels = 1:nlevels(data[[section_column]]))
##         gap = c(rep(1, nlevels(data[[section_column]]) - 1), 
##             start_gap)
##         axis_min <- round(min(data[[histogram_column]]), 3)
##         axis_min <- round(axis_min + (axis_min * 0.1), 3)
##         axis_min_half <- round(axis_min/2, 3)
##         axis_max <- round(max(data[[histogram_column]]), 3)
##         axis_max <- round(axis_max + (axis_max * 0.1), 3)
##         axis_max_half <- round(axis_max/2, 3)
##         circlize::circos.trackHist(factors = data$section_numbers, 
##             x = data[[histogram_column]], track.height = track3_height, 
##             track.index = NULL, col = lines_col3, border = lines_col3, 
##             bg.border = NA, draw.density = histogram_densityplot, 
##             bin.size = histogram_binsize)
##         circlize::circos.yaxis(side = y_axis_location, sector.index = x_axis_index, 
##             track.index = track.index, at = c(axis_min, track_axis_reference, 
##                 axis_max), tick = y_axis_tick, tick.length = y_axis_tick_length, 
##             labels.cex = y_axis_label_cex)
##     }
##     if (legend == TRUE && track_number == 1) {
##         legend1 <- ComplexHeatmap::Legend(at = c(track1_label), 
##             labels_gp = grid::gpar(fontsize = 15), ncol = 1, 
##             border = NA, background = NA, legend_gp = grid::gpar(col = c(discrete_palette[1])), 
##             type = "points", pch = 19, size = grid::unit(15, 
##                 "mm"), grid_height = grid::unit(15, "mm"), grid_width = grid::unit(15, 
##                 "mm"), direction = "vertical")
##     }
##     if (legend == TRUE && track_number >= 2) {
##         legend1 <- ComplexHeatmap::Legend(at = c(track1_label, 
##             track2_label), labels_gp = grid::gpar(fontsize = 15), 
##             ncol = 1, border = NA, background = NA, legend_gp = grid::gpar(col = c(discrete_palette[1], 
##                 discrete_palette[2])), type = "points", pch = 19, 
##             size = grid::unit(15, "mm"), grid_height = grid::unit(15, 
##                 "mm"), grid_width = grid::unit(15, "mm"), direction = "vertical")
##     }
##     if (legend == TRUE && track_number >= 3) {
##         legend1 <- ComplexHeatmap::Legend(at = c(track1_label, 
##             track2_label, track3_label), labels_gp = grid::gpar(fontsize = 15), 
##             ncol = 1, border = NA, background = NA, legend_gp = grid::gpar(col = c(discrete_palette[1], 
##                 discrete_palette[2], discrete_palette[3])), type = "points", 
##             pch = 19, size = grid::unit(15, "mm"), grid_height = grid::unit(15, 
##                 "mm"), grid_width = grid::unit(15, "mm"), direction = "vertical")
##     }
##     if (legend == TRUE) {
##         legend2 <- ComplexHeatmap::Legend(at = pvalue_label, 
##             labels_gp = grid::gpar(fontsize = 15), ncol = 1, 
##             border = NA, background = NA, legend_gp = grid::gpar(col = c("black")), 
##             type = "points", pch = 1, size = grid::unit(15, "mm"), 
##             grid_height = grid::unit(15, "mm"), grid_width = grid::unit(15, 
##                 "mm"), direction = "vertical")
##         names <- levels(as.factor(data[[section_column]]))
##         names <- paste(1:nlevels(data[[section_column]]), names, 
##             sep = ". ")
##         legend3 <- ComplexHeatmap::Legend(at = names, labels_gp = grid::gpar(fontsize = 15), 
##             nrow = 4, ncol = 7, border = NA, background = NA, 
##             legend_gp = grid::gpar(col = c("black")), size = grid::unit(15, 
##                 "mm"), grid_height = grid::unit(15, "mm"), grid_width = grid::unit(10, 
##                 "mm"), direction = "horizontal")
##         legend4 <- ComplexHeatmap::packLegend(legend1, legend2, 
##             direction = "vertical", gap = grid::unit(0, "mm"))
##         legend <- ComplexHeatmap::packLegend(legend4, legend3, 
##             direction = "horizontal", gap = grid::unit(0, "mm"))
##         legend_height <- legend@grob[["vp"]][["height"]]
##         legend_width <- legend@grob[["vp"]][["width"]]
##         grid::pushViewport(grid::viewport(x = grid::unit(0.5, 
##             "npc"), y = grid::unit(0.08, "npc"), width = legend_width, 
##             height = legend_height, just = c("center", "top")))
##         grid::grid.draw(legend)
##         grid::upViewport()
##     }
## }
## <bytecode: 0x7fb9b2fcd9a8>
## <environment: namespace:EpiCircos>

Chapter 6: R package code

library(EpiCircos)
circos_plot(track_number = 3,
            track1_data = EpiCircos::EpiCircos_data,
            track2_data = EpiCircos::EpiCircos_data,
            track3_data = EpiCircos::EpiCircos_data,
            track1_type = "points", track2_type = "lines", track3_type = "bar",
            label_column = 1, section_column = 2,
            estimate_column = 4, pvalue_column = 5,
            pvalue_adjustment = 1,
            lower_ci = 7, upper_ci = 8,
            lines_column = 10, lines_type = "o",
            bar_column = 9,
            legend = TRUE,
            track1_label = "Track 1",
            track2_label = "Track 2",
            track3_label = "Track 3",
            pvalue_label = "<= 0.05",
            circle_size = 25)

Chapter 6: application

Chapter 7 and 8: overview

Clustering metabolites / rules for instrumenting clusters

Chapter (0%)

Chapters will compare a number of different approaches to metabolite clustering from the literature and set out guidance for instrumenting these clusters

• Priors

class; subclass; biological pathway; size; shared genetic variants

• No priors

PCA; factor analysis; hierarchical clustering; density clustering; self organising map; LDSR; ontology

• Focus
  • SOM: distance between metabolites (k-means clustering) - Mika/Ville
  • Factor analysis: PCA but identify latent variable - Wes/Kate
  • Ontology: vectorise metabolites and identify distances - Ben Elsworth
• Which betas do you use if same SNP multiple times in a cluster?

Chapter 9

MR analysis: metabolites to diseases

Chapter (5%): writing (0%), analysis (20%)

Paper (0%): combined with chapter 5 / stand alone paper

Chapter explores association between metabolites and diseases

• metabolites
  • chapters 5 and 7
• diseases
  • systematic review
• code tested - scaleable
  • 1-2 weeks to complete analysis
• how to present
  • up-to 123 metabolites
  • potentially > 100 diseases
  • database / network graph / subsets…

Chapter 10

Discussion/limitations/conclusion

Chapter (2%): writing (0%), analysis (20%)

Chaper to present:

Analysis pipeline figure

Timeline

Acknowledgements

• Kaitlin Wade
• Laura Corbin
• Nic Timpson
• David Hughes
• Charlie Hatcher
• Nancy McBride
• Steph Suddell
• Tom Battram
• Luke McGuinness
• Caroline Bull
• Osama Mahmoud